Self-Driving Car Engineer Nanodegree¶
Deep Learning¶
Project: Build a Traffic Sign Recognition Classifier¶
In this notebook, a template is provided for you to implement your functionality in stages which is required to successfully complete this project. If additional code is required that cannot be included in the notebook, be sure that the Python code is successfully imported and included in your submission, if necessary. Sections that begin with 'Implementation' in the header indicate where you should begin your implementation for your project. Note that some sections of implementation are optional, and will be marked with 'Optional' in the header.
In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.
Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. In addition, Markdown cells can be edited by typically double-clicking the cell to enter edit mode.
Step 0: Load The Data¶
# Load pickled data
import pickle
import numpy as np
from pathlib import Path
# TODO: Fill this in based on where you saved the training and testing data
training_file = "/home/hsharma/Downloads/UND/traffic-signs-data/train.p"
testing_file = "/home/hsharma/Downloads/UND/traffic-signs-data/test.p"
with open(training_file, mode='rb') as f:
train = pickle.load(f)
with open(testing_file, mode='rb') as f:
test = pickle.load(f)
X_train, y_train = train['features'], train['labels']
X_test, y_test = test['features'], test['labels']
class_labels = np.genfromtxt('signnames.csv', dtype=None, delimiter=',', skip_header=1)
print ("data loaded")
is_data_flip_extended = Path("extended_flip.pickle").is_file()
is_data_transform_extended = Path("extended_dataset3.pickle").is_file()
is_data_preprocessed = Path('final_train.pickle').is_file() and Path('final_valid.pickle').is_file()
is_data_trimmed = Path("trimmed_data.pickle").is_file()
print ("is_data_flip_extended = ", is_data_flip_extended)
print ("is_data_transform_extended = ", is_data_transform_extended)
print ("is_data_preprocessed = ", is_data_preprocessed)
print ("is_data_trimmed = ", is_data_trimmed)
Step 1: Dataset Summary & Exploration¶
The pickled data is a dictionary with 4 key/value pairs:
'features'is a 4D array containing raw pixel data of the traffic sign images, (num examples, width, height, channels).'labels'is a 2D array containing the label/class id of the traffic sign. The filesignnames.csvcontains id -> name mappings for each id.'sizes'is a list containing tuples, (width, height) representing the the original width and height the image.'coords'is a list containing tuples, (x1, y1, x2, y2) representing coordinates of a bounding box around the sign in the image. THESE COORDINATES ASSUME THE ORIGINAL IMAGE. THE PICKLED DATA CONTAINS RESIZED VERSIONS (32 by 32) OF THESE IMAGES
Complete the basic data summary below.
### Replace each question mark with the appropriate value.
# TODO: Number of training examples
n_train = X_train.shape[0]
# TODO: Number of testing examples.
n_test = X_test.shape[0]
# TODO: What's the shape of an traffic sign image?
image_shape = str(X_train.shape[1])+"x"+str(X_train.shape[2])+"x"+str(X_train.shape[3])
# TODO: How many unique classes/labels there are in the dataset.
n_classes = len(np.unique(y_train))
print("Number of training examples =", n_train)
print("Number of testing examples =", n_test)
print("Image data shape =", image_shape)
print("Number of classes =", n_classes)
Visualize the German Traffic Signs Dataset using the pickled file(s). This is open ended, suggestions include: plotting traffic sign images, plotting the count of each sign, etc.
The Matplotlib examples and gallery pages are a great resource for doing visualizations in Python.
NOTE: It's recommended you start with something simple first. If you wish to do more, come back to it after you've completed the rest of the sections.
# Define a function to classify traiing classes based on number of samples to help during augmentation
def count_images(counter):
augment_level_1 = []
augment_level_2 = []
augment_level_3 = []
augment_level_4 = []
augment_level_5 = []
for i in range(len(counter)):
if counter[i] <= 300:
augment_level_5.append(i)
elif counter[i] <= 500:
augment_level_4.append(i)
elif counter[i] <= 900:
augment_level_3.append(i)
elif counter[i] <= 1500:
augment_level_2.append(i)
else:
augment_level_1.append(i)
return augment_level_1, augment_level_2, augment_level_3, augment_level_4, augment_level_5
### Data exploration visualization goes here.
### Feel free to use as many code cells as needed.
import matplotlib.pyplot as plt
from random import randint
# Visualizations will be shown in the notebook.
%matplotlib inline
# Show a random image from dataset along with the label and class
image_num = randint(0,n_train)
print ("Showing image number ", image_num)
plt.figure(1)
plt.title(str(class_labels[y_train[image_num]][0])+" = "+str(class_labels[y_train[image_num]][1]))
plt.imshow(X_train[image_num])
# Count of each class
unique, counter = np.unique(y_train, return_counts=True)
print ("class count ", counter)
# Visualize the unbalance in the classes
f2 = plt.figure(2)
plt.hist(y_train, bins=n_classes)
f2.show()
### Preprocess the data here.
### Feel free to use as many code cells as needed.
def save_data_pickle(fname, x, y):
# print('Saving data to pickle file...')
try:
with open(fname, 'wb') as pfile:
pickle.dump(
{
'X_train_extended': x,
"y_train_extended": y
},
pfile, pickle.HIGHEST_PROTOCOL)
except Exception as e:
# print('Unable to save data to', fname, ':', e)
raise
# print('Data cached in pickle file.')
## Generate new data by flipping images
if not is_data_flip_extended:
flip_y_axis = np.array([11, 18, 22, 26, 30, 35])
flip_x_axis = np.array([5])
flip_both = np.array([15, 17])
flip_origin = np.array([32, 40])
X_train_extended = np.empty([0, X_train.shape[1], X_train.shape[2], X_train.shape[3]], dtype = X_train.dtype)
y_train_extended = np.empty([0], dtype = y_train.dtype)
i = 0
while (i < len(y_train)):
if y_train[i] in flip_y_axis:
X_train_extended = np.append(X_train_extended, [X_train[i]], axis=0)
X_train_extended = np.append(X_train_extended, [X_train[i, :,::-1,:]], axis=0)
y_train_extended = np.append(y_train_extended, y_train[i])
y_train_extended = np.append(y_train_extended, y_train[i])
i+=1
if y_train[i] in flip_x_axis:
X_train_extended = np.append(X_train_extended, [X_train[i]], axis=0)
X_train_extended = np.append(X_train_extended, [X_train[i, ::-1, :, :]], axis=0)
y_train_extended = np.append(y_train_extended, y_train[i])
y_train_extended = np.append(y_train_extended, y_train[i])
i+=1
if y_train[i] in flip_both:
X_train_extended = np.append(X_train_extended, [X_train[i]], axis=0)
X_train_extended = np.append(X_train_extended, [X_train[i, ::-1, :, :]], axis=0)
X_train_extended = np.append(X_train_extended, [X_train[i, :,::-1,:]], axis=0)
y_train_extended = np.append(y_train_extended, y_train[i])
y_train_extended = np.append(y_train_extended, y_train[i])
y_train_extended = np.append(y_train_extended, y_train[i])
i+=1
if y_train[i] in flip_origin:
X_train_extended = np.append(X_train_extended, [X_train[i]], axis=0)
X_train_extended = np.append(X_train_extended, [X_train[i, ::-1, ::-1, :]], axis=0)
y_train_extended = np.append(y_train_extended, y_train[i])
y_train_extended = np.append(y_train_extended, y_train[i])
i+=1
if not (y_train[i] in flip_y_axis or y_train[i] in flip_both or
y_train[i] in flip_x_axis or y_train[i] in flip_origin):
X_train_extended = np.append(X_train_extended, X_train[i:i+counter[y_train[i]]], axis=0)
y_train_extended = np.append(y_train_extended, y_train[i:i+counter[y_train[i]]])
i+=counter[y_train[i]]
save_data_pickle('extended_flip.pickle', X_train_extended, y_train_extended)
is_data_flip_extended = True
Checkpoint 1¶
Save data
with open('extended_flip.pickle', mode='rb') as f:
train = pickle.load(f)
X_train_extended = train['X_train_extended']
y_train_extended = train['y_train_extended']
unique, counter = np.unique(y_train_extended, return_counts=True)
augment_level_1, augment_level_2, augment_level_3, augment_level_4, augment_level_5 = count_images(counter)
print ("class count ", counter)
f2 = plt.figure(2)
plt.hist(y_train_extended, bins=n_classes)
f2.show()
from skimage.transform import rotate
from skimage.transform import warp
from skimage.transform import AffineTransform
from skimage.transform import ProjectiveTransform
from skimage import img_as_ubyte
import random
import sys
if not is_data_transform_extended:
def augment_image(image, level):
x_size = 32
y_size = 32
img1 = image
# Rotated
img_rot_arr = []
for i in range(level):
img_rot_arr.append(img_as_ubyte(rotate(img1,random.uniform(-20,20), mode='edge')))
# Translated
img_translated_arr = []
tform = AffineTransform(translation=[random.uniform(-4,4),random.uniform(-4,4)])
img_translated_arr.append(img_as_ubyte(warp(img1, tform)))
for i in range(len(img_rot_arr)):
tform = AffineTransform(translation=[random.uniform(-4,4),random.uniform(-4,4)])
img_translated_arr.append(img_as_ubyte(warp(img_rot_arr[i], tform)))
# Sheared
tform1 = AffineTransform(shear=random.uniform(-0.2, 0))
tform2 = AffineTransform(shear=random.uniform(0.01,0.21))
img_neg_shear_arr = []
img_pos_shear_arr = []
img_rev1_shear_arr = []
img_rev2_shear_arr = []
img_neg_shear_arr.append(img_as_ubyte(warp(img1, tform1)))
img_pos_shear_arr.append(img_as_ubyte(warp(img1, tform2)))
img_rev1_shear_arr.append(img_as_ubyte(warp(img_neg_shear_arr[0], tform2)))
img_rev2_shear_arr.append(img_as_ubyte(warp(img_pos_shear_arr[0], tform1)))
for i in range(len(img_rot_arr)):
tform1 = AffineTransform(shear=random.uniform(-0.2, 0))
tform2 = AffineTransform(shear=random.uniform(0.01,0.21))
img_neg_shear_arr.append(img_as_ubyte(warp(img_rot_arr[i], tform1)))
img_pos_shear_arr.append(img_as_ubyte(warp(img_rot_arr[i], tform2)))
img_rev1_shear_arr.append(img_as_ubyte(warp(img_neg_shear_arr[i+1], tform2)))
img_rev2_shear_arr.append(img_as_ubyte(warp(img_pos_shear_arr[i+1], tform1)))
# Projective transform
transform = ProjectiveTransform()
dst = np.array(((0, 0),(0, y_size),(x_size, y_size),(x_size, 0)))
img_proj_trans_arr = []
var = random.uniform(-5,5)
src = (np.array(((var, var),(var, y_size-var),(x_size - var, y_size - var),(x_size - var, var))))
transform.estimate(src, dst)
img_proj_trans_arr.append(img_as_ubyte(warp(img1, transform, order = 1, mode = 'edge')))
for i in range(len(img_rot_arr)):
var = random.uniform(-5,5)
src = (np.array(((var, var),(var, y_size-var),(x_size - var, y_size - var),(x_size - var, var))))
transform.estimate(src, dst)
img_proj_trans_arr.append(img_as_ubyte(warp(img_rot_arr[i], transform, order = 1, mode = 'edge')))
img_proj_trans_arr.append(img_as_ubyte(warp(img_translated_arr[i], transform, order = 1, mode = 'edge')))
if level in [1]:
final_arr = [img1] + img_rot_arr + img_translated_arr + img_pos_shear_arr
elif level in [2,3]:
final_arr = [img1] + img_rot_arr + img_translated_arr + img_pos_shear_arr + img_proj_trans_arr
else:
final_arr = [img1] + img_rot_arr + img_translated_arr + img_neg_shear_arr + img_pos_shear_arr + img_rev1_shear_arr + img_rev2_shear_arr + img_proj_trans_arr
return final_arr
# X_train_extended2 = np.empty([0, X_train.shape[1], X_train.shape[2], X_train.shape[3]], dtype = X_train.dtype)
# y_train_extended2 = np.empty([0], dtype = y_train.dtype)
ss = 1000;
total = X_train_extended.shape[0]
# i = 0
while i < total:
if y_train_extended[i] in augment_level_5:
images = augment_image(X_train_extended[i], 6)
elif y_train_extended[i] in augment_level_4:
images = augment_image(X_train_extended[i], 4)
elif y_train_extended[i] in augment_level_3:
images = augment_image(X_train_extended[i], 3)
elif y_train_extended[i] in augment_level_2:
images = augment_image(X_train_extended[i], 2)
elif y_train_extended[i] in augment_level_1:
images = augment_image(X_train_extended[i], 1)
X_train_extended2 = np.append(X_train_extended2,images, axis=0)
y_train_extended2 = np.append(y_train_extended2,[y_train_extended[i]]*len(images))
if (i%ss == 0):
sys.stdout.write("i = "+str(i))
i += 1
sys.stdout.write(str(X_train_extended2.shape))
sys.stdout.write(str(y_train_extended2.shape))
save_data_pickle('extended_dataset3.pickle', X_train_extended2, y_train_extended2)
is_data_transform_extended = True
Checkpoint 2¶
Save extended dataset
# Visualize final dataset
# save_data_pickle('extended_dataset_t.pickle', X_train_extended2, y_train_extended2)
with open('/home/hsharma/UND/CarND-Traffic-Sign-Classifier-Project/extended_dataset3.pickle', mode='rb') as f:
train = pickle.load(f)
X_train_final = train['X_train_extended']
y_train_final = train['y_train_extended']
# # print (j)
# X_train_final = X_train_extended2
# y_train_final = y_train_extended2
unique, counter2 = np.unique(y_train_final, return_counts=True)
plt.figure(2)
plt.hist(y_train_final, bins=n_classes)
plt.show()
if not is_data_trimmed:
i = 0
X_trimmed = np.empty([0, X_train.shape[1], X_train.shape[2], X_train.shape[3]], dtype = X_train.dtype)
y_trimmed = np.empty([0], dtype = y_train.dtype)
start = 0
while (i<n_classes):
print (i)
end = start+min(8000, counter2[i])
X_trimmed = np.append(X_trimmed, X_train_final[start:end], axis=0)
y_trimmed = np.append(y_trimmed, y_train_final[start:end])
start += counter2[i]
i += 1
print (X_trimmed.shape)
print (y_trimmed.shape)
save_data_pickle('trimmed_data.pickle', X_trimmed, y_trimmed)
is_data_trimmed = False
if is_data_trimmed:
with open('trimmed_data.pickle', mode='rb') as f:
train = pickle.load(f)
X_trimmed = train['X_train_extended']
y_trimmed = train['y_train_extended']
if not is_data_trimmed:
X_trimmed = X_train_final
y_trimmed = y_train_final
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelBinarizer
from sklearn.utils import shuffle
import tensorflow
from skimage import color
if not is_data_preprocessed:
# Apply One-Hot Encoding
encoder = LabelBinarizer()
encoder.fit(y_trimmed)
y_train_final = encoder.transform(y_trimmed)
y_test = encoder.transform(y_test)
# Change to float32, so it can be multiplied against the features in TensorFlow, which are float32
y_train_final = y_train_final.astype(np.float32)
y_test = y_test.astype(np.float32)
# Convert to grayscale
X_train_gray = np.empty([0, 32,32], dtype=X_train_final.dtype)
chunk_size = 100000
for i in range(0, X_train_final.shape[0], chunk_size):
start = i
# fname = "exg"+str(i)+".pickle"
print (start)
end = min(i+chunk_size, X_train_final.shape[0])
X_train_gray = np.append(X_train_gray, np.dot(X_trimmed[start:end][...,:3],[0.299,0.587,0.114]), axis=0)
# save_data_pickle(f_name, X_train_gray, y_train_final[start:end])
X_test_gray = np.dot(X_test[...][...,:3],[0.299,0.587,0.114])
# Normalize
X_train_g = (X_train_gray - np.mean(X_train_gray))/np.std(X_train_gray)
X_test_g =(X_test_gray - np.mean(X_test_gray))/np.std(X_test_gray)
# Split into training and validation
train_features, valid_features, train_labels, valid_labels = train_test_split(
X_train_g,
y_train_final,
test_size=0.20,
random_state=8998)
# Final save
save_data_pickle('final_train.pickle', train_features, train_labels)
save_data_pickle('final_valid.pickle', valid_features, valid_labels)
save_data_pickle('final_test.pickle', X_test_g, y_test)
Checkpoint 3¶
Finished preprocessing
train_features, valid_features, train_labels, valid_labels = train_test_split(
X_trimmed,
y_trimmed,
test_size=0.23,
random_state=8998)
# Shuffle
train_features, train_labels = shuffle(train_features, train_labels)
valid_features, valid_labels = shuffle(valid_features, valid_labels)
import random
col_width = max(len(label) for label in class_labels)
sign_classes, class_indices, class_counts = np.unique(y_trimmed, return_index = True, return_counts = True)
for c, c_index, c_count in zip(sign_classes, class_indices, class_counts):
print("Class %i: %-*s %s samples" % (c, col_width, class_labels[c], str(c_count)))
fig = plt.figure(figsize = (6, 1))
fig.subplots_adjust(left = 0, right = 1, bottom = 0, top = 1, hspace = 0.05, wspace = 0.05)
random_indices = random.sample(range(c_index, c_index + c_count), 10)
for i in range(10):
axis = fig.add_subplot(1, 10, i + 1, xticks=[], yticks=[])
axis.imshow(X_trimmed[random_indices[i]])
plt.show()
print("--------------------------------------------------------------------------------------\n")
Question 1¶
Describe how you preprocessed the data. Why did you choose that technique?
Answer:
On visualizing the number of images in each classes, I realized the data was quite skewed. So, in order to make the dataset uniform, I used data augmentation
Question 2¶
Describe how you set up the training, validation and testing data for your model. Optional: If you generated additional data, how did you generate the data? Why did you generate the data? What are the differences in the new dataset (with generated data) from the original dataset?
Answer:
For data augmentation I followed the following steps
1) Identified classes that can be flipped either along the horizontal axis or the vertical axis or the origin. I considered only the classes which do not change on flipping. There were some classes for which the label needed to be updated after flipping, so I ignored these.
2) Created new np arrays containing the original and flipped images and the respective classes
3) Since, even after flip the data was still not uniform, I added more images using the augment_image function which adds the images with following transforms
- Roatated between angle -20 to 20
- Translated between -3 to 3 on both the axis
- Sheared with a factor between -0.2 to 0
- Sheared with a factor between 0.01 to 0.2
- Sheared c,d with factor of opposite sign as earlier
- Projective transformed
4) To make all the data uniform and remove help improve accuracy, I augmented all classes so that there are no issues because of the black border added due to translation and shearing
5) Once the data was augmented and saved, I used 20% of it for validation and remaining for training
Step 2: Design and Test a Model Architecture¶
Design and implement a deep learning model that learns to recognize traffic signs. Train and test your model on the German Traffic Sign Dataset.
There are various aspects to consider when thinking about this problem:
- Neural network architecture
- Play around preprocessing techniques (normalization, rgb to grayscale, etc)
- Number of examples per label (some have more than others).
- Generate fake data.
Here is an example of a published baseline model on this problem. It's not required to be familiar with the approach used in the paper but, it's good practice to try to read papers like these.
NOTE: The LeNet-5 implementation shown in the classroom at the end of the CNN lesson is a solid starting point. You'll have to change the number of classes and possibly the preprocessing, but aside from that it's plug and play!
Implementation¶
Use the code cell (or multiple code cells, if necessary) to implement the first step of your project. Once you have completed your implementation and are satisfied with the results, be sure to thoroughly answer the questions that follow.
### Define your architecture here.
### Feel free to use as many code cells as needed.
import tensorflow as tf
from tensorflow.contrib.layers import flatten
def LeNet(x):
mu = 0
sigma = 0.1
# SOLUTION: Layer 1: Convolutional. Input = 32x32x3. Output = 28x28x6.
conv1_W = tf.Variable(tf.truncated_normal(shape=(5, 5, 3, 6), mean = mu, stddev = sigma))
conv1_b = tf.Variable(tf.zeros(6))
conv1 = tf.nn.conv2d(x, conv1_W, strides=[1, 1, 1, 1], padding='VALID') + conv1_b
# SOLUTION: Activation.
conv1 = tf.nn.relu(conv1)
# SOLUTION: Pooling. Input = 28x28x6. Output = 14x14x6.
conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
# SOLUTION: Layer 2: Convolutional. Output = 10x10x16.
conv2_W = tf.Variable(tf.truncated_normal(shape=(5, 5, 6, 16), mean = mu, stddev = sigma))
conv2_b = tf.Variable(tf.zeros(16))
conv2 = tf.nn.conv2d(conv1, conv2_W, strides=[1, 1, 1, 1], padding='VALID') + conv2_b
# SOLUTION: Activation.
conv2 = tf.nn.relu(conv2)
# SOLUTION: Pooling. Input = 10x10x16. Output = 5x5x16.
conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
# SOLUTION: Flatten. Input = 5x5x16. Output = 400.
fc0 = flatten(conv2)
# SOLUTION: Layer 3: Fully Connected. Input = 400. Output = 120.
fc1_W = tf.Variable(tf.truncated_normal(shape=(400, 120), mean = mu, stddev = sigma))
fc1_b = tf.Variable(tf.zeros(120))
fc1 = tf.matmul(fc0, fc1_W) + fc1_b
# SOLUTION: Activation.
fc1 = tf.nn.relu(fc1)
# SOLUTION: Layer 4: Fully Connected. Input = 120. Output = 84.
fc2_W = tf.Variable(tf.truncated_normal(shape=(120, 84), mean = mu, stddev = sigma))
fc2_b = tf.Variable(tf.zeros(84))
fc2 = tf.matmul(fc1, fc2_W) + fc2_b
# SOLUTION: Activation.
fc2 = tf.nn.relu(fc2)
# SOLUTION: Layer 5: Fully Connected. Input = 84. Output = 43.
fc3_W = tf.Variable(tf.truncated_normal(shape=(84, 43), mean = mu, stddev = sigma))
fc3_b = tf.Variable(tf.zeros(43))
logits = tf.matmul(fc2, fc3_W) + fc3_b
return logits
def LeNetModified(x):
mu = 0
sigma = 0.11
dropout = 0.75
# SOLUTION: Layer 1: Convolutional. Input = 32x32x3. Output = 30x30x8.
conv1_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 3, 8), mean = mu, stddev = sigma))
conv1_b = tf.Variable(tf.zeros(8))
conv1 = tf.nn.conv2d(x, conv1_W, strides=[1, 1, 1, 1], padding='VALID') + conv1_b
# SOLUTION: Activation.
conv1 = tf.nn.relu(conv1)
# SOLUTION: Pooling. Input = 30x30x8. Output = 15x15x8.
conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
#----------------------------------------------------------------------------------
# SOLUTION: Layer 2: Convolutional. Inout = 15x15x8. Output = 12x12x32.
conv2_W = tf.Variable(tf.truncated_normal(shape=(4, 4, 8, 32), mean = mu, stddev = sigma))
conv2_b = tf.Variable(tf.zeros(32))
conv2 = tf.nn.conv2d(conv1, conv2_W, strides=[1, 1, 1, 1], padding='VALID') + conv2_b
# SOLUTION: Activation.
conv2 = tf.nn.relu(conv2)
# conv2 = tf.nn.tanh(conv2)
# SOLUTION: Pooling. Input = 12x12x32. Output = 6x6x32.
conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
#----------------------------------------------------------------------------------
# SOLUTION: Flatten. Input = 6x6x32. Output = 1152.
fc0 = flatten(conv2)
#----------------------------------------------------------------------------------
# SOLUTION: Layer 4: Fully Connected. Input = 600. Output = 450.
fc1_W = tf.Variable(tf.truncated_normal(shape=(1152, 576), mean = mu, stddev = sigma))
fc1_b = tf.Variable(tf.zeros(576))
fc1 = tf.matmul(fc0, fc1_W) + fc1_b
# SOLUTION: Activation.
fc1 = tf.nn.relu(fc1)
#----------------------------------------------------------------------------------
# SOLUTION: Layer 5: Fully Connected. Input = 450. Output = 300.
fc2_W = tf.Variable(tf.truncated_normal(shape=(576, 192), mean = mu, stddev = sigma))
fc2_b = tf.Variable(tf.zeros(192))
fc2 = tf.matmul(fc1, fc2_W) + fc2_b
# SOLUTION: Activation.
fc2 = tf.nn.relu(fc2)
#----------------------------------------------------------------------------------
# SOLUTION: Layer 5: Fully Connected. Input = 300. Output = 150.
fc3_W = tf.Variable(tf.truncated_normal(shape=(192, 96), mean = mu, stddev = sigma))
fc3_b = tf.Variable(tf.zeros(96))
fc3 = tf.matmul(fc2, fc3_W) + fc3_b
# SOLUTION: Activation.
fc3 = tf.nn.relu(fc3)
#----------------------------------------------------------------------------------
# SOLUTION: Layer 5: Fully Connected. Input = 150. Output = 43.
fc4_W = tf.Variable(tf.truncated_normal(shape=(96, 43), mean = mu, stddev = sigma))
fc4_b = tf.Variable(tf.zeros(43))
logits = tf.matmul(fc3, fc4_W) + fc4_b
return logits
Question 3¶
What does your final architecture look like? (Type of model, layers, sizes, connectivity, etc.) For reference on how to build a deep neural network using TensorFlow, see Deep Neural Network in TensorFlow from the classroom.
Answer: After playing around with the number of layers, the activation functions, dropout and max pooling size, the final architecture follows a LeNet model
Layer 1: Convolutional.¶
Input = 32x32x3. Output = 28x28x6.
Activation - relu
max_pooling (k = 2) Output - 14x14x6
Layer 2: Convolutional.¶
Input = 14x14x6. Output = 10x10x16.
Activation - relu
max_pooling (k = 2) Output - 5x5x16
Flatten.¶
- Input = 5x5x16. Output = 400.
Layer 3: Fully Connected.¶
Input = 400. Output = 120.
Activation - relu
Layer 4: Fully Connected.¶
Input = 120. Output = 84.
Activation - relu
Layer 5: Fully Connected.¶
- Input = 84. Output = 43.
from tqdm import tqdm
EPOCHS = 15
BATCH_SIZE = 256
rate = 0.001
x = tf.placeholder(tf.float32, (None, 32, 32,3))
y = tf.placeholder(tf.int32, (None))
one_hot_y = tf.one_hot(y, 43)
logits = LeNet(x)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits, one_hot_y)
loss_operation = tf.reduce_mean(cross_entropy)
optimizer = tf.train.AdamOptimizer(learning_rate = rate)
training_operation = optimizer.minimize(loss_operation)
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(one_hot_y, 1))
accuracy_operation = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
saver = tf.train.Saver()
def evaluate(X_data, y_data):
num_examples = len(X_data)
total_accuracy = 0
sess = tf.get_default_session()
for offset in range(0, num_examples, BATCH_SIZE):
batch_x, batch_y = X_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
accuracy = sess.run(accuracy_operation, feed_dict={x: batch_x, y: batch_y})
total_accuracy += (accuracy * len(batch_x))
return total_accuracy / num_examples
EPOCHS = 50
BATCH_SIZE = 420
import time
val_acc = [];
start_time = time.time()
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
num_examples = len(X_train_final)
print("Training...")
print()
for i in range(EPOCHS):
print (i+1, " of ", EPOCHS)
train_features, train_labels = shuffle(X_train_final, y_train_final)
for offset in tqdm(range(0, num_examples, BATCH_SIZE)):
end = offset + BATCH_SIZE
batch_x, batch_y = train_features[offset:end], train_labels[offset:end]
sess.run(training_operation, feed_dict={x: batch_x, y: batch_y})
# train_accuracy = evaluate(train_features, train_labels)
validation_accuracy = evaluate(valid_features, valid_labels)
print("EPOCH {} ...".format(i+1))
# print("Training Accuracy = {:.3f}".format(train_accuracy))
print("Validation Accuracy = {:.3f}".format(validation_accuracy))
val_acc.append(validation_accuracy)
print()
saver.save(sess, './models/lenet__2')
print("Model saved")
print ("Time taken = ", (time.time()-start_time)/60, " minutes")
with tf.Session() as sess:
saver.restore(sess, tf.train.latest_checkpoint('.'))
test_accuracy = evaluate(X_test, y_test)
print("Test Accuracy = {:.3f}".format(test_accuracy))
Question 4¶
How did you train your model? (Type of optimizer, batch size, epochs, hyperparameters, etc.)
Answer:
Training:Validation - 80:20
Optimizer - AdamOptimizer
Batch Size - 420
Epochs - 50
Learning Rate - 0.001
Softmax on logits
Question 5¶
What approach did you take in coming up with a solution to this problem? It may have been a process of trial and error, in which case, outline the steps you took to get to the final solution and why you chose those steps. Perhaps your solution involved an already well known implementation or architecture. In this case, discuss why you think this is suitable for the current problem.
Answer:
I started by impleneting a solution using the LeNet architecture and trained it for a small number of eopchs (15). The accuracy was around 97. After that I played around with the architecture, added a few convolutional layers, changed the size of filters, used combinations of sigmoid, relu and tanh for activation, added additional fully connected layers, increased epochs, varied batch size, added dropout etc.
The maximum I achieved was 98.5 with 2 convolutonal layer one with relu activation and anther with tanh and both having max pooling with k=2 followed by 4 fully connected layers, second one having tanh and others relu, trained for 35 epochs on a batch size of 400.
In order to keep the architecture simple, I trained LeNet for 50 epochs and batch size of 420 and since there was not much difference with the modified architecture, I proceeded with LeNet
Step 3: Test a Model on New Images¶
Take several pictures of traffic signs that you find on the web or around you (at least five), and run them through your classifier on your computer to produce example results. The classifier might not recognize some local signs but it could prove interesting nonetheless.
You may find signnames.csv useful as it contains mappings from the class id (integer) to the actual sign name.
Implementation¶
Use the code cell (or multiple code cells, if necessary) to implement the first step of your project. Once you have completed your implementation and are satisfied with the results, be sure to thoroughly answer the questions that follow.
import os
import matplotlib.pyplot as plt
import numpy as np
from PIL import Image
# img.thumbnail((64, 64), Image.ANTIALIAS) # resizes image in-place
# imgplot = plt.imshow(img)
folder_name = "new_images/"
img_arr = []
labels_actual = []
for fname in os.listdir(folder_name):
print (fname)
labels_actual.append(int(fname[:-4]))
img = Image.open(folder_name + fname)
img_resized = img.resize((32,32))
img_resized = np.asarray(img_resized)
img_arr.append(img_resized)
fig = plt.figure(figsize = (2, 1))
# Show original image
axis = fig.add_subplot(1, 2, 1, xticks=[], yticks=[])
axis.imshow(img)
# Show resized image
axis = fig.add_subplot(1, 2, 2, xticks=[], yticks=[])
axis.imshow(img_resized)
plt.show()
print("----------------------------------------------\n")
new_images = np.array(img_arr)
new_images_labels_actual = np.array(labels_actual)
print (new_images_labels_actual)
print (new_images.shape)
import tensorflow as tf
saver = tf.train.Saver()
with tf.Session() as sess:
saver.restore(sess, tf.train.latest_checkpoint('.'))
new_images_labels_predicted = sess.run(logits, feed_dict={x: new_images})
test_accuracy = evaluate(new_images, new_images_labels_actual)
print("Test Accuracy = {:.3f}".format(test_accuracy))
top7s = []
for i in range(0, new_images_labels_predicted.shape[0],1):
with tf.Session() as sess:
top7s.append(sess.run(tf.nn.top_k(tf.constant(new_images_labels_predicted[i]), k=7)) )
y_axis = range(0,7,1)
for i in range(len(top7s)):
top7 = top7s[i][1]
top7_vals = top7s[i][0]
top7_labels = [str(class_labels[x]) for x in top7]
norm2 = [x/np.sum(top7_vals) for x in top7_vals]
color_val = "black"
actual_class = new_images_labels_actual[i]
if top7[0] == actual_class:
color_val = "green"
fig, ax = plt.subplots()
rects = ax.bar(y_axis, norm2, width=0.5, color=color_val)
ax.set_title("Image "+str(i))
ax.set_xlabel("Actual class = "+str(class_labels[actual_class]))
ax.set_ylabel("Confidence")
ax.set_xticklabels(top7_labels, size="small", rotation=60)
# Show original image
fig = plt.figure(figsize = (2, 1))
axis = fig.add_subplot(1, 2, 1, xticks=[], yticks=[])
axis.imshow(new_images[i])
# Show resized image
k = np.sum(counter2[:actual_class])+random.randint(1,counter2[actual_class]-1)
axis = fig.add_subplot(1, 2, 2, xticks=[], yticks=[])
axis.imshow(X_train_final[k])
print ("Top predictions = ", top7)
print ("Actual class = ", class_labels[actual_class])
i+=1
print ("--------------------")
Question 6¶
Choose five candidate images of traffic signs and provide them in the report. Are there any particular qualities of the image(s) that might make classification difficult? It could be helpful to plot the images in the notebook.
Answer:
I identified the following issues that led to incorrect classification of images
- Scaling down the images to 32x32 distorts the pixels (Image 3)
- The shape of the sign also affects the classfication (Image 5)
- The differences in backgroud colors in training and test (Image 13)
### Run the predictions here.
### Feel free to use as many code cells as needed.
Question 7¶
Is your model able to perform equally well on captured pictures when compared to testing on the dataset? The simplest way to do this check the accuracy of the predictions. For example, if the model predicted 1 out of 5 signs correctly, it's 20% accurate.
NOTE: You could check the accuracy manually by using signnames.csv (same directory). This file has a mapping from the class id (0-42) to the corresponding sign name. So, you could take the class id the model outputs, lookup the name in signnames.csv and see if it matches the sign from the image.
Answer:
The model was not able to classify the images that well mainly because of the factore listed in Q6 but if we ignore those datapoints, the classification performed reasonably well
Question 8¶
Use the model's softmax probabilities to visualize the certainty of its predictions, tf.nn.top_k could prove helpful here. Which predictions is the model certain of? Uncertain? If the model was incorrect in its initial prediction, does the correct prediction appear in the top k? (k should be 5 at most)
tf.nn.top_k will return the values and indices (class ids) of the top k predictions. So if k=3, for each sign, it'll return the 3 largest probabilities (out of a possible 43) and the correspoding class ids.
Take this numpy array as an example:
# (5, 6) array
a = np.array([[ 0.24879643, 0.07032244, 0.12641572, 0.34763842, 0.07893497,
0.12789202],
[ 0.28086119, 0.27569815, 0.08594638, 0.0178669 , 0.18063401,
0.15899337],
[ 0.26076848, 0.23664738, 0.08020603, 0.07001922, 0.1134371 ,
0.23892179],
[ 0.11943333, 0.29198961, 0.02605103, 0.26234032, 0.1351348 ,
0.16505091],
[ 0.09561176, 0.34396535, 0.0643941 , 0.16240774, 0.24206137,
0.09155967]])Running it through sess.run(tf.nn.top_k(tf.constant(a), k=3)) produces:
TopKV2(values=array([[ 0.34763842, 0.24879643, 0.12789202],
[ 0.28086119, 0.27569815, 0.18063401],
[ 0.26076848, 0.23892179, 0.23664738],
[ 0.29198961, 0.26234032, 0.16505091],
[ 0.34396535, 0.24206137, 0.16240774]]), indices=array([[3, 0, 5],
[0, 1, 4],
[0, 5, 1],
[1, 3, 5],
[1, 4, 3]], dtype=int32))Looking just at the first row we get [ 0.34763842, 0.24879643, 0.12789202], you can confirm these are the 3 largest probabilities in a. You'll also notice [3, 0, 5] are the corresponding indices.
Answer:
Visualized above.¶
Note: Once you have completed all of the code implementations and successfully answered each question above, you may finalize your work by exporting the iPython Notebook as an HTML document. You can do this by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.